[
https://issues.apache.org/jira/browse/FLINK-15093?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16989609#comment-16989609
]
Kurt Young commented on FLINK-15093:
------------------------------------
We also met similar problem in sql client: FLINK-15052
I think the reason is we didn't clear the buffered transformation in
`StreamExecutionEnvironment`. Currently, both sql client and
`StreamExecutionEnvironment#execute()` will get `StreamGraph` first and try to
execute the `StreamGraph`.
Does it make sense to have an extra boolean flag to indicate whether we want to
also clear transformations when getting `StreamGraph`? Just like
`ExecutionEnvironment#createProgramPlan(String jobName, boolean clearSinks)`
did.
> FLINK-13708 still exist in StreamExecutionEnvironment
> -----------------------------------------------------
>
> Key: FLINK-15093
> URL: https://issues.apache.org/jira/browse/FLINK-15093
> Project: Flink
> Issue Type: Bug
> Components: API / DataStream
> Affects Versions: 1.10.0, 1.9.1
> Reporter: Jeff Zhang
> Priority: Critical
> Attachments: screenshot-1.png
>
>
> Use the following code in scala shell to reproduce this issue.
> {code}
> val data = senv.fromElements("hello world", "hello flink", "hello hadoop")
> data.flatMap(line => line.split("\\s")).
> map(w => (w, 1)).
> keyBy(0).
> sum(1).
> print
> senv.execute()
> data.flatMap(line => line.split("\\s")).
> map(w => (w, 1)).
> keyBy(0).
> sum(1).
> print
> senv.execute()
> {code}
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
